Notice: The reproducibility variables underlying each score are classified using an automated LLM-based pipeline, validated against a manually labeled dataset. LLM-based classification introduces uncertainty and potential bias; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in [1].

Mixture Martingales Revisited with Applications to Sequential Tests and Confidence Intervals

Authors: Emilie Kaufmann, Wouter M. Koolen

JMLR 2021 | Venue PDF | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We illustrate this empirically in Figure 4 for a Bernoulli bandit model with M arms with mean 0.1 and 4 more arms with means 0.2, 0.3, 0.4, 0.5 (thus K = M + 4), for different values of M. We consider the use of a Box weight vector that is uniform on the singletons (π(1) = 1), a weight vector supported on the whole set of arms (π(K) = 1) and a weight vector that is uniform over subset sizes (π(k) = 1/K). For each value of M, data is collected using uniform sampling and we set δ = 10^-10 to focus on the high confidence regime.
Researcher Affiliation Academia Emilie Kaufmann Univ. Lille, CNRS, Inria, Centrale Lille, UMR 9189 CRISt AL, F-59000 Lille, France EMAIL; Wouter M. Koolen Centrum Wiskunde & Informatica, Science Park 123, Amsterdam, Netherlands EMAIL
Pseudocode No The paper describes algorithms and methods verbally, such as the 'Tracking rule' and 'GLR stopping rule', but does not provide any explicitly labeled pseudocode blocks or algorithms in a structured, code-like format.
Open Source Code No The paper does not contain any explicit statement about releasing source code for the described methodology, nor does it provide a link to a code repository.
Open Datasets No The paper discusses various types of distributions (e.g., Gaussian, Bernoulli, Gamma) for multi-armed bandit models and uses a 'Bernoulli bandit model' for empirical illustration. However, it does not provide access information (link, DOI, specific citation to an external dataset with authors/year) for any publicly available or open dataset. The empirical illustration uses simulated data based on specific mean values, not a pre-existing dataset.
Dataset Splits No The paper does not use external datasets and therefore does not provide any training/test/validation dataset split information.
Hardware Specification No The paper does not mention any specific hardware (e.g., GPU models, CPU types, or cloud computing specifications) used for running its experiments or simulations.
Software Dependencies Yes For both Gaussian and Bernoulli (and possibly more) we can write the objective as a Disciplined Convex Program and solve it efficiently with e.g. CVX (Grant and Boyd, 2017). CVX: Matlab software for disciplined convex programming, version 2.1. http://cvxr.com/cvx, Mar. 2017.
Experiment Setup Yes We illustrate this empirically in Figure 4 for a Bernoulli bandit model with M arms with mean 0.1 and 4 more arms with means 0.2, 0.3, 0.4, 0.5 (thus K = M + 4), for different values of M. We consider the use of a Box weight vector that is uniform on the singletons (π(1) = 1), a weight vector supported on the whole set of arms (π(K) = 1) and a weight vector that is uniform over subset sizes (π(k) = 1/K). For each value of M, data is collected using uniform sampling and we set δ = 10^-10 to focus on the high confidence regime.